Efficiency of Automated Detectors of Learner Engagement and Affect Compared With Traditional Observation Methods

• In Event: Poster Session 3
  In Poster Session: Exploring the Current Landscape of Technology, Instruction, Cognition, and Learning

Abstract

Student engagement with learning activities has consistently been tied to academic achievement. To positively influence learning outcomes, malleable factors must be measured and strategies devised to improve them. We tested the hypothesis that automated detectors of affect and engagement are expensive to develop but that, if they can be applied at scale to the log files of many learners, they will produce and process observation data more cost-efficiently than traditional observation methods such as in-class observations and video analysis. We compare the reliability and costs of collecting engagement and affect data using automated detectors with three other methods and conclude that while indeed vastly more cost-efficient when applied at scale, automated detectors are not yet as reliable as in-person observations

Authors

• Fiona M. Hollands, Teachers College, Columbia University

• Ipek Bakir, Teachers College, Columbia University